datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2015 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "datapath.h"
  56 #include "conntrack.h"
  57 #include "flow.h"
  58 #include "flow_table.h"
  59 #include "flow_netlink.h"
  60 #include "gso.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66 EXPORT_SYMBOL_GPL(ovs_net_id);
  67
  68 static struct genl_family dp_packet_genl_family;
  69 static struct genl_family dp_flow_genl_family;
  70 static struct genl_family dp_datapath_genl_family;
  71
  72 static const struct nla_policy flow_policy[];
  73
  74 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  75         .name = OVS_FLOW_MCGROUP
  76 };
  77
  78 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  79         .name = OVS_DATAPATH_MCGROUP
  80 };
  81
  82 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  83         .name = OVS_VPORT_MCGROUP
  84 };
  85
  86 /* Check if need to build a reply message.
  87  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply.
  88  */
  89 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  90                             unsigned int group)
  91 {
  92         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  93                genl_has_listeners(family, genl_info_net(info), group);
  94 }
  95
  96 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  97                        struct sk_buff *skb, struct genl_info *info)
  98 {
  99         genl_notify(family, skb, genl_info_net(info),
 100                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
 101 }
 102
 103 /**
 104  * DOC: Locking:
 105  *
 106  * All writes e.g. Writes to device state (add/remove datapath, port, set
 107  * operations on vports, etc.), Writes to other state (flow table
 108  * modifications, set miscellaneous datapath parameters, etc.) are protected
 109  * by ovs_lock.
 110  *
 111  * Reads are protected by RCU.
 112  *
 113  * There are a few special cases (mostly stats) that have their own
 114  * synchronization but they nest under all of above and don't interact with
 115  * each other.
 116  *
 117  * The RTNL lock nests inside ovs_mutex.
 118  */
 119
 120 static DEFINE_MUTEX(ovs_mutex);
 121
 122 void ovs_lock(void)
 123 {
 124         mutex_lock(&ovs_mutex);
 125 }
 126
 127 void ovs_unlock(void)
 128 {
 129         mutex_unlock(&ovs_mutex);
 130 }
 131
 132 #ifdef CONFIG_LOCKDEP
 133 int lockdep_ovsl_is_held(void)
 134 {
 135         if (debug_locks)
 136                 return lockdep_is_held(&ovs_mutex);
 137         else
 138                 return 1;
 139 }
 140 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 141 #endif
 142
 143 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 144                              const struct sw_flow_key *,
 145                              const struct dp_upcall_info *);
 146 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 147                                   const struct sw_flow_key *,
 148                                   const struct dp_upcall_info *);
 149
 150 /* Must be called with rcu_read_lock. */
 151 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 152 {
 153         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 154
 155         if (dev) {
 156                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 157                 if (vport)
 158                         return vport->dp;
 159         }
 160
 161         return NULL;
 162 }
 163
 164 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 165  * returned dp pointer valid.
 166  */
 167 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 168 {
 169         struct datapath *dp;
 170
 171         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 172         rcu_read_lock();
 173         dp = get_dp_rcu(net, dp_ifindex);
 174         rcu_read_unlock();
 175
 176         return dp;
 177 }
 178
 179 /* Must be called with rcu_read_lock or ovs_mutex. */
 180 const char *ovs_dp_name(const struct datapath *dp)
 181 {
 182         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 183         return ovs_vport_name(vport);
 184 }
 185
 186 static int get_dpifindex(const struct datapath *dp)
 187 {
 188         struct vport *local;
 189         int ifindex;
 190
 191         rcu_read_lock();
 192
 193         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 194         if (local)
 195                 ifindex = local->dev->ifindex;
 196         else
 197                 ifindex = 0;
 198
 199         rcu_read_unlock();
 200
 201         return ifindex;
 202 }
 203
 204 static void destroy_dp_rcu(struct rcu_head *rcu)
 205 {
 206         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 207
 208         ovs_flow_tbl_destroy(&dp->table);
 209         free_percpu(dp->stats_percpu);
 210         kfree(dp->ports);
 211         kfree(dp);
 212 }
 213
 214 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 215                                             u16 port_no)
 216 {
 217         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 218 }
 219
 220 /* Called with ovs_mutex or RCU read lock. */
 221 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 222 {
 223         struct vport *vport;
 224         struct hlist_head *head;
 225
 226         head = vport_hash_bucket(dp, port_no);
 227         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 228                 if (vport->port_no == port_no)
 229                         return vport;
 230         }
 231         return NULL;
 232 }
 233
 234 /* Called with ovs_mutex. */
 235 static struct vport *new_vport(const struct vport_parms *parms)
 236 {
 237         struct vport *vport;
 238
 239         vport = ovs_vport_add(parms);
 240         if (!IS_ERR(vport)) {
 241                 struct datapath *dp = parms->dp;
 242                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 243
 244                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 245         }
 246         return vport;
 247 }
 248
 249 void ovs_dp_detach_port(struct vport *p)
 250 {
 251         ASSERT_OVSL();
 252
 253         /* First drop references to device. */
 254         hlist_del_rcu(&p->dp_hash_node);
 255
 256         /* Then destroy it. */
 257         ovs_vport_del(p);
 258 }
 259
 260 /* Must be called with rcu_read_lock. */
 261 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 262 {
 263         const struct vport *p = OVS_CB(skb)->input_vport;
 264         struct datapath *dp = p->dp;
 265         struct sw_flow *flow;
 266         struct sw_flow_actions *sf_acts;
 267         struct dp_stats_percpu *stats;
 268         u64 *stats_counter;
 269         u32 n_mask_hit;
 270
 271         stats = this_cpu_ptr(dp->stats_percpu);
 272
 273         /* Look up flow. */
 274         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 275                                          &n_mask_hit);
 276         if (unlikely(!flow)) {
 277                 struct dp_upcall_info upcall;
 278                 int error;
 279
 280                 memset(&upcall, 0, sizeof(upcall));
 281                 upcall.cmd = OVS_PACKET_CMD_MISS;
 282                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 283                 upcall.mru = OVS_CB(skb)->mru;
 284                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 285                 if (unlikely(error))
 286                         kfree_skb(skb);
 287                 else
 288                         consume_skb(skb);
 289                 stats_counter = &stats->n_missed;
 290                 goto out;
 291         }
 292
 293         ovs_flow_stats_update(flow, key->tp.flags, skb);
 294         sf_acts = rcu_dereference(flow->sf_acts);
 295         ovs_execute_actions(dp, skb, sf_acts, key);
 296
 297         stats_counter = &stats->n_hit;
 298
 299 out:
 300         /* Update datapath statistics. */
 301         u64_stats_update_begin(&stats->syncp);
 302         (*stats_counter)++;
 303         stats->n_mask_hit += n_mask_hit;
 304         u64_stats_update_end(&stats->syncp);
 305 }
 306
 307 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 308                   const struct sw_flow_key *key,
 309                   const struct dp_upcall_info *upcall_info)
 310 {
 311         struct dp_stats_percpu *stats;
 312         int err;
 313
 314         if (upcall_info->portid == 0) {
 315                 err = -ENOTCONN;
 316                 goto err;
 317         }
 318
 319         if (!skb_is_gso(skb))
 320                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 321         else
 322                 err = queue_gso_packets(dp, skb, key, upcall_info);
 323         if (err)
 324                 goto err;
 325
 326         return 0;
 327
 328 err:
 329         stats = this_cpu_ptr(dp->stats_percpu);
 330
 331         u64_stats_update_begin(&stats->syncp);
 332         stats->n_lost++;
 333         u64_stats_update_end(&stats->syncp);
 334
 335         return err;
 336 }
 337
 338 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 339                              const struct sw_flow_key *key,
 340                              const struct dp_upcall_info *upcall_info)
 341 {
 342         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 343         struct sw_flow_key later_key;
 344         struct sk_buff *segs, *nskb;
 345         struct ovs_skb_cb ovs_cb;
 346         int err;
 347
 348         ovs_cb = *OVS_CB(skb);
 349         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 350         *OVS_CB(skb) = ovs_cb;
 351         if (IS_ERR(segs))
 352                 return PTR_ERR(segs);
 353         if (segs == NULL)
 354                 return -EINVAL;
 355
 356         if (gso_type & SKB_GSO_UDP) {
 357                 /* The initial flow key extracted by ovs_flow_key_extract()
 358                  * in this case is for a first fragment, so we need to
 359                  * properly mark later fragments.
 360                  */
 361                 later_key = *key;
 362                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 363         }
 364
 365         /* Queue all of the segments. */
 366         skb = segs;
 367         do {
 368                 *OVS_CB(skb) = ovs_cb;
 369                 if (gso_type & SKB_GSO_UDP && skb != segs)
 370                         key = &later_key;
 371
 372                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 373                 if (err)
 374                         break;
 375
 376         } while ((skb = skb->next));
 377
 378         /* Free all of the segments. */
 379         skb = segs;
 380         do {
 381                 nskb = skb->next;
 382                 if (err)
 383                         kfree_skb(skb);
 384                 else
 385                         consume_skb(skb);
 386         } while ((skb = nskb));
 387         return err;
 388 }
 389
 390 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 391                               unsigned int hdrlen)
 392 {
 393         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 394                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 395                 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 396
 397         /* OVS_PACKET_ATTR_USERDATA */
 398         if (upcall_info->userdata)
 399                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 400
 401         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 402         if (upcall_info->egress_tun_info)
 403                 size += nla_total_size(ovs_tun_key_attr_size());
 404
 405         /* OVS_PACKET_ATTR_ACTIONS */
 406         if (upcall_info->actions_len)
 407                 size += nla_total_size(upcall_info->actions_len);
 408
 409         /* OVS_PACKET_ATTR_MRU */
 410         if (upcall_info->mru)
 411                 size += nla_total_size(sizeof(upcall_info->mru));
 412
 413         return size;
 414 }
 415
 416 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 417 {
 418         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 419                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 420
 421                 if (plen > 0)
 422                         memset(skb_put(skb, plen), 0, plen);
 423         }
 424 }
 425
 426 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 427                                   const struct sw_flow_key *key,
 428                                   const struct dp_upcall_info *upcall_info)
 429 {
 430         struct ovs_header *upcall;
 431         struct sk_buff *nskb = NULL;
 432         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 433         struct nlattr *nla;
 434         struct genl_info info = {
 435 #ifdef HAVE_GENLMSG_NEW_UNICAST
 436                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 437 #endif
 438                 .snd_portid = upcall_info->portid,
 439         };
 440         size_t len;
 441         unsigned int hlen;
 442         int err, dp_ifindex;
 443
 444         dp_ifindex = get_dpifindex(dp);
 445         if (!dp_ifindex)
 446                 return -ENODEV;
 447
 448         if (skb_vlan_tag_present(skb)) {
 449                 nskb = skb_clone(skb, GFP_ATOMIC);
 450                 if (!nskb)
 451                         return -ENOMEM;
 452
 453                 nskb = vlan_insert_tag_set_proto(nskb, nskb->vlan_proto, skb_vlan_tag_get(nskb));
 454                 if (!nskb)
 455                         return -ENOMEM;
 456
 457                 vlan_set_tci(nskb, 0);
 458
 459                 skb = nskb;
 460         }
 461
 462         if (nla_attr_size(skb->len) > USHRT_MAX) {
 463                 err = -EFBIG;
 464                 goto out;
 465         }
 466
 467         /* Complete checksum if needed */
 468         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 469             (err = skb_checksum_help(skb)))
 470                 goto out;
 471
 472         /* Older versions of OVS user space enforce alignment of the last
 473          * Netlink attribute to NLA_ALIGNTO which would require extensive
 474          * padding logic. Only perform zerocopy if padding is not required.
 475          */
 476         if (dp->user_features & OVS_DP_F_UNALIGNED)
 477                 hlen = skb_zerocopy_headlen(skb);
 478         else
 479                 hlen = skb->len;
 480
 481         len = upcall_msg_size(upcall_info, hlen);
 482         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 483         if (!user_skb) {
 484                 err = -ENOMEM;
 485                 goto out;
 486         }
 487
 488         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 489                              0, upcall_info->cmd);
 490         upcall->dp_ifindex = dp_ifindex;
 491
 492         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 493         BUG_ON(err);
 494
 495         if (upcall_info->userdata)
 496                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 497                           nla_len(upcall_info->userdata),
 498                           nla_data(upcall_info->userdata));
 499
 500
 501         if (upcall_info->egress_tun_info) {
 502                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 503                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 504                                                     upcall_info->egress_tun_info,
 505                                                     upcall_info->egress_tun_opts);
 506                 BUG_ON(err);
 507                 nla_nest_end(user_skb, nla);
 508         }
 509
 510         if (upcall_info->actions_len) {
 511                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 512                 err = ovs_nla_put_actions(upcall_info->actions,
 513                                           upcall_info->actions_len,
 514                                           user_skb);
 515                 if (!err)
 516                         nla_nest_end(user_skb, nla);
 517                 else
 518                         nla_nest_cancel(user_skb, nla);
 519         }
 520
 521         /* Add OVS_PACKET_ATTR_MRU */
 522         if (upcall_info->mru) {
 523                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 524                                 upcall_info->mru)) {
 525                         err = -ENOBUFS;
 526                         goto out;
 527                 }
 528                 pad_packet(dp, user_skb);
 529         }
 530
 531         /* Only reserve room for attribute header, packet data is added
 532          * in skb_zerocopy()
 533          */
 534         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 535                 err = -ENOBUFS;
 536                 goto out;
 537         }
 538         nla->nla_len = nla_attr_size(skb->len);
 539
 540         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 541         if (err)
 542                 goto out;
 543
 544         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 545         pad_packet(dp, user_skb);
 546
 547         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 548
 549         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 550         user_skb = NULL;
 551 out:
 552         if (err)
 553                 skb_tx_error(skb);
 554         kfree_skb(user_skb);
 555         kfree_skb(nskb);
 556         return err;
 557 }
 558
 559 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 560 {
 561         struct ovs_header *ovs_header = info->userhdr;
 562         struct net *net = sock_net(skb->sk);
 563         struct nlattr **a = info->attrs;
 564         struct sw_flow_actions *acts;
 565         struct sk_buff *packet;
 566         struct sw_flow *flow;
 567         struct sw_flow_actions *sf_acts;
 568         struct datapath *dp;
 569         struct ethhdr *eth;
 570         struct vport *input_vport;
 571         u16 mru = 0;
 572         int len;
 573         int err;
 574         bool log = !a[OVS_PACKET_ATTR_PROBE];
 575
 576         err = -EINVAL;
 577         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 578             !a[OVS_PACKET_ATTR_ACTIONS])
 579                 goto err;
 580
 581         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 582         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 583         err = -ENOMEM;
 584         if (!packet)
 585                 goto err;
 586         skb_reserve(packet, NET_IP_ALIGN);
 587
 588         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 589
 590         skb_reset_mac_header(packet);
 591         eth = eth_hdr(packet);
 592
 593         /* Normally, setting the skb 'protocol' field would be handled by a
 594          * call to eth_type_trans(), but it assumes there's a sending
 595          * device, which we may not have.
 596          */
 597         if (eth_proto_is_802_3(eth->h_proto))
 598                 packet->protocol = eth->h_proto;
 599         else
 600                 packet->protocol = htons(ETH_P_802_2);
 601
 602         /* Set packet's mru */
 603         if (a[OVS_PACKET_ATTR_MRU]) {
 604                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 605                 packet->ignore_df = 1;
 606         }
 607         OVS_CB(packet)->mru = mru;
 608
 609         /* Build an sw_flow for sending this packet. */
 610         flow = ovs_flow_alloc();
 611         err = PTR_ERR(flow);
 612         if (IS_ERR(flow))
 613                 goto err_kfree_skb;
 614
 615         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 616                                              packet, &flow->key, log);
 617         if (err)
 618                 goto err_flow_free;
 619
 620         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 621                                    &flow->key, &acts, log);
 622         if (err)
 623                 goto err_flow_free;
 624
 625         rcu_assign_pointer(flow->sf_acts, acts);
 626         packet->priority = flow->key.phy.priority;
 627         packet->mark = flow->key.phy.skb_mark;
 628
 629         rcu_read_lock();
 630         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 631         err = -ENODEV;
 632         if (!dp)
 633                 goto err_unlock;
 634
 635         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 636         if (!input_vport)
 637                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 638
 639         if (!input_vport)
 640                 goto err_unlock;
 641
 642         packet->dev = input_vport->dev;
 643         OVS_CB(packet)->input_vport = input_vport;
 644         sf_acts = rcu_dereference(flow->sf_acts);
 645
 646         local_bh_disable();
 647         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 648         local_bh_enable();
 649         rcu_read_unlock();
 650
 651         ovs_flow_free(flow, false);
 652         return err;
 653
 654 err_unlock:
 655         rcu_read_unlock();
 656 err_flow_free:
 657         ovs_flow_free(flow, false);
 658 err_kfree_skb:
 659         kfree_skb(packet);
 660 err:
 661         return err;
 662 }
 663
 664 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 665         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 666         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 667         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 668         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 669         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 670 };
 671
 672 static struct genl_ops dp_packet_genl_ops[] = {
 673         { .cmd = OVS_PACKET_CMD_EXECUTE,
 674           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 675           .policy = packet_policy,
 676           .doit = ovs_packet_cmd_execute
 677         }
 678 };
 679
 680 static struct genl_family dp_packet_genl_family = {
 681         .id = GENL_ID_GENERATE,
 682         .hdrsize = sizeof(struct ovs_header),
 683         .name = OVS_PACKET_FAMILY,
 684         .version = OVS_PACKET_VERSION,
 685         .maxattr = OVS_PACKET_ATTR_MAX,
 686         .netnsok = true,
 687         .parallel_ops = true,
 688         .ops = dp_packet_genl_ops,
 689         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 690 };
 691
 692 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 693                          struct ovs_dp_megaflow_stats *mega_stats)
 694 {
 695         int i;
 696
 697         memset(mega_stats, 0, sizeof(*mega_stats));
 698
 699         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 700         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 701
 702         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 703
 704         for_each_possible_cpu(i) {
 705                 const struct dp_stats_percpu *percpu_stats;
 706                 struct dp_stats_percpu local_stats;
 707                 unsigned int start;
 708
 709                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 710
 711                 do {
 712                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 713                         local_stats = *percpu_stats;
 714                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 715
 716                 stats->n_hit += local_stats.n_hit;
 717                 stats->n_missed += local_stats.n_missed;
 718                 stats->n_lost += local_stats.n_lost;
 719                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 720         }
 721 }
 722
 723 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 724 {
 725         return ovs_identifier_is_ufid(sfid) &&
 726                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 727 }
 728
 729 static bool should_fill_mask(uint32_t ufid_flags)
 730 {
 731         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 732 }
 733
 734 static bool should_fill_actions(uint32_t ufid_flags)
 735 {
 736         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 737 }
 738
 739 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 740                                     const struct sw_flow_id *sfid,
 741                                     uint32_t ufid_flags)
 742 {
 743         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 744
 745         /* OVS_FLOW_ATTR_UFID */
 746         if (sfid && ovs_identifier_is_ufid(sfid))
 747                 len += nla_total_size(sfid->ufid_len);
 748
 749         /* OVS_FLOW_ATTR_KEY */
 750         if (!sfid || should_fill_key(sfid, ufid_flags))
 751                 len += nla_total_size(ovs_key_attr_size());
 752
 753         /* OVS_FLOW_ATTR_MASK */
 754         if (should_fill_mask(ufid_flags))
 755                 len += nla_total_size(ovs_key_attr_size());
 756
 757         /* OVS_FLOW_ATTR_ACTIONS */
 758         if (should_fill_actions(ufid_flags))
 759                 len += nla_total_size(acts->orig_len);
 760
 761         return len
 762                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 763                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 764                 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 765 }
 766
 767 /* Called with ovs_mutex or RCU read lock. */
 768 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 769                                    struct sk_buff *skb)
 770 {
 771         struct ovs_flow_stats stats;
 772         __be16 tcp_flags;
 773         unsigned long used;
 774
 775         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 776
 777         if (used &&
 778             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 779                 return -EMSGSIZE;
 780
 781         if (stats.n_packets &&
 782             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 783                 return -EMSGSIZE;
 784
 785         if ((u8)ntohs(tcp_flags) &&
 786              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 787                 return -EMSGSIZE;
 788
 789         return 0;
 790 }
 791
 792 /* Called with ovs_mutex or RCU read lock. */
 793 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 794                                      struct sk_buff *skb, int skb_orig_len)
 795 {
 796         struct nlattr *start;
 797         int err;
 798
 799         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 800          * this is the first flow to be dumped into 'skb'.  This is unusual for
 801          * Netlink but individual action lists can be longer than
 802          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 803          * The userspace caller can always fetch the actions separately if it
 804          * really wants them.  (Most userspace callers in fact don't care.)
 805          *
 806          * This can only fail for dump operations because the skb is always
 807          * properly sized for single flows.
 808          */
 809         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 810         if (start) {
 811                 const struct sw_flow_actions *sf_acts;
 812
 813                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 814                 err = ovs_nla_put_actions(sf_acts->actions,
 815                                           sf_acts->actions_len, skb);
 816
 817                 if (!err)
 818                         nla_nest_end(skb, start);
 819                 else {
 820                         if (skb_orig_len)
 821                                 return err;
 822
 823                         nla_nest_cancel(skb, start);
 824                 }
 825         } else if (skb_orig_len) {
 826                 return -EMSGSIZE;
 827         }
 828
 829         return 0;
 830 }
 831
 832 /* Called with ovs_mutex or RCU read lock. */
 833 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 834                                   struct sk_buff *skb, u32 portid,
 835                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 836 {
 837         const int skb_orig_len = skb->len;
 838         struct ovs_header *ovs_header;
 839         int err;
 840
 841         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 842                                  flags, cmd);
 843         if (!ovs_header)
 844                 return -EMSGSIZE;
 845
 846         ovs_header->dp_ifindex = dp_ifindex;
 847
 848         err = ovs_nla_put_identifier(flow, skb);
 849         if (err)
 850                 goto error;
 851
 852         if (should_fill_key(&flow->id, ufid_flags)) {
 853                 err = ovs_nla_put_masked_key(flow, skb);
 854                 if (err)
 855                         goto error;
 856         }
 857
 858         if (should_fill_mask(ufid_flags)) {
 859                 err = ovs_nla_put_mask(flow, skb);
 860                 if (err)
 861                         goto error;
 862         }
 863
 864         err = ovs_flow_cmd_fill_stats(flow, skb);
 865         if (err)
 866                 goto error;
 867
 868         if (should_fill_actions(ufid_flags)) {
 869                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 870                 if (err)
 871                         goto error;
 872         }
 873
 874         genlmsg_end(skb, ovs_header);
 875         return 0;
 876
 877 error:
 878         genlmsg_cancel(skb, ovs_header);
 879         return err;
 880 }
 881
 882 /* May not be called with RCU read lock. */
 883 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 884                                                const struct sw_flow_id *sfid,
 885                                                struct genl_info *info,
 886                                                bool always,
 887                                                uint32_t ufid_flags)
 888 {
 889         struct sk_buff *skb;
 890         size_t len;
 891
 892         if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
 893                                         GROUP_ID(&ovs_dp_flow_multicast_group)))
 894                 return NULL;
 895
 896         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 897         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 898         if (!skb)
 899                 return ERR_PTR(-ENOMEM);
 900
 901         return skb;
 902 }
 903
 904 /* Called with ovs_mutex. */
 905 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 906                                                int dp_ifindex,
 907                                                struct genl_info *info, u8 cmd,
 908                                                bool always, u32 ufid_flags)
 909 {
 910         struct sk_buff *skb;
 911         int retval;
 912
 913         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 914                                       &flow->id, info, always, ufid_flags);
 915         if (IS_ERR_OR_NULL(skb))
 916                 return skb;
 917
 918         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 919                                         info->snd_portid, info->snd_seq, 0,
 920                                         cmd, ufid_flags);
 921         BUG_ON(retval < 0);
 922         return skb;
 923 }
 924
 925 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 926 {
 927         struct net *net = sock_net(skb->sk);
 928         struct nlattr **a = info->attrs;
 929         struct ovs_header *ovs_header = info->userhdr;
 930         struct sw_flow *flow = NULL, *new_flow;
 931         struct sw_flow_mask mask;
 932         struct sk_buff *reply;
 933         struct datapath *dp;
 934         struct sw_flow_key key;
 935         struct sw_flow_actions *acts;
 936         struct sw_flow_match match;
 937         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 938         int error;
 939         bool log = !a[OVS_FLOW_ATTR_PROBE];
 940
 941         /* Must have key and actions. */
 942         error = -EINVAL;
 943         if (!a[OVS_FLOW_ATTR_KEY]) {
 944                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 945                 goto error;
 946         }
 947         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 948                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 949                 goto error;
 950         }
 951
 952         /* Most of the time we need to allocate a new flow, do it before
 953          * locking.
 954          */
 955         new_flow = ovs_flow_alloc();
 956         if (IS_ERR(new_flow)) {
 957                 error = PTR_ERR(new_flow);
 958                 goto error;
 959         }
 960
 961         /* Extract key. */
 962         ovs_match_init(&match, &key, &mask);
 963         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 964                                   a[OVS_FLOW_ATTR_MASK], log);
 965         if (error)
 966                 goto err_kfree_flow;
 967
 968         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 969
 970         /* Extract flow identifier. */
 971         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 972                                        &key, log);
 973         if (error)
 974                 goto err_kfree_flow;
 975
 976         /* Validate actions. */
 977         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 978                                      &new_flow->key, &acts, log);
 979         if (error) {
 980                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 981                 goto err_kfree_flow;
 982         }
 983
 984         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 985                                         ufid_flags);
 986         if (IS_ERR(reply)) {
 987                 error = PTR_ERR(reply);
 988                 goto err_kfree_acts;
 989         }
 990
 991         ovs_lock();
 992         dp = get_dp(net, ovs_header->dp_ifindex);
 993         if (unlikely(!dp)) {
 994                 error = -ENODEV;
 995                 goto err_unlock_ovs;
 996         }
 997
 998         /* Check if this is a duplicate flow */
 999         if (ovs_identifier_is_ufid(&new_flow->id))
1000                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1001         if (!flow)
1002                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1003         if (likely(!flow)) {
1004                 rcu_assign_pointer(new_flow->sf_acts, acts);
1005
1006                 /* Put flow in bucket. */
1007                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1008                 if (unlikely(error)) {
1009                         acts = NULL;
1010                         goto err_unlock_ovs;
1011                 }
1012
1013                 if (unlikely(reply)) {
1014                         error = ovs_flow_cmd_fill_info(new_flow,
1015                                                        ovs_header->dp_ifindex,
1016                                                        reply, info->snd_portid,
1017                                                        info->snd_seq, 0,
1018                                                        OVS_FLOW_CMD_NEW,
1019                                                        ufid_flags);
1020                         BUG_ON(error < 0);
1021                 }
1022                 ovs_unlock();
1023         } else {
1024                 struct sw_flow_actions *old_acts;
1025
1026                 /* Bail out if we're not allowed to modify an existing flow.
1027                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1028                  * because Generic Netlink treats the latter as a dump
1029                  * request.  We also accept NLM_F_EXCL in case that bug ever
1030                  * gets fixed.
1031                  */
1032                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1033                                                          | NLM_F_EXCL))) {
1034                         error = -EEXIST;
1035                         goto err_unlock_ovs;
1036                 }
1037                 /* The flow identifier has to be the same for flow updates.
1038                  * Look for any overlapping flow.
1039                  */
1040                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1041                         if (ovs_identifier_is_key(&flow->id))
1042                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1043                                                                  &match);
1044                         else /* UFID matches but key is different */
1045                                 flow = NULL;
1046                         if (!flow) {
1047                                 error = -ENOENT;
1048                                 goto err_unlock_ovs;
1049                         }
1050                 }
1051                 /* Update actions. */
1052                 old_acts = ovsl_dereference(flow->sf_acts);
1053                 rcu_assign_pointer(flow->sf_acts, acts);
1054
1055                 if (unlikely(reply)) {
1056                         error = ovs_flow_cmd_fill_info(flow,
1057                                                        ovs_header->dp_ifindex,
1058                                                        reply, info->snd_portid,
1059                                                        info->snd_seq, 0,
1060                                                        OVS_FLOW_CMD_NEW,
1061                                                        ufid_flags);
1062                         BUG_ON(error < 0);
1063                 }
1064                 ovs_unlock();
1065
1066                 ovs_nla_free_flow_actions_rcu(old_acts);
1067                 ovs_flow_free(new_flow, false);
1068         }
1069
1070         if (reply)
1071                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1072         return 0;
1073
1074 err_unlock_ovs:
1075         ovs_unlock();
1076         kfree_skb(reply);
1077 err_kfree_acts:
1078         ovs_nla_free_flow_actions(acts);
1079 err_kfree_flow:
1080         ovs_flow_free(new_flow, false);
1081 error:
1082         return error;
1083 }
1084
1085 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1086 static struct sw_flow_actions *get_flow_actions(struct net *net,
1087                                                 const struct nlattr *a,
1088                                                 const struct sw_flow_key *key,
1089                                                 const struct sw_flow_mask *mask,
1090                                                 bool log)
1091 {
1092         struct sw_flow_actions *acts;
1093         struct sw_flow_key masked_key;
1094         int error;
1095
1096         ovs_flow_mask_key(&masked_key, key, true, mask);
1097         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1098         if (error) {
1099                 OVS_NLERR(log,
1100                           "Actions may not be safe on all matching packets");
1101                 return ERR_PTR(error);
1102         }
1103
1104         return acts;
1105 }
1106
1107 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1108 {
1109         struct net *net = sock_net(skb->sk);
1110         struct nlattr **a = info->attrs;
1111         struct ovs_header *ovs_header = info->userhdr;
1112         struct sw_flow_key key;
1113         struct sw_flow *flow;
1114         struct sw_flow_mask mask;
1115         struct sk_buff *reply = NULL;
1116         struct datapath *dp;
1117         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1118         struct sw_flow_match match;
1119         struct sw_flow_id sfid;
1120         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1121         int error;
1122         bool log = !a[OVS_FLOW_ATTR_PROBE];
1123         bool ufid_present;
1124
1125         /* Extract key. */
1126         error = -EINVAL;
1127         if (!a[OVS_FLOW_ATTR_KEY]) {
1128                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1129                 goto error;
1130         }
1131
1132         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1133         ovs_match_init(&match, &key, &mask);
1134         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1135                                   a[OVS_FLOW_ATTR_MASK], log);
1136         if (error)
1137                 goto error;
1138
1139         /* Validate actions. */
1140         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1141                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1142                                         &mask, log);
1143                 if (IS_ERR(acts)) {
1144                         error = PTR_ERR(acts);
1145                         goto error;
1146                 }
1147
1148                 /* Can allocate before locking if have acts. */
1149                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1150                                                 ufid_flags);
1151                 if (IS_ERR(reply)) {
1152                         error = PTR_ERR(reply);
1153                         goto err_kfree_acts;
1154                 }
1155         }
1156
1157         ovs_lock();
1158         dp = get_dp(net, ovs_header->dp_ifindex);
1159         if (unlikely(!dp)) {
1160                 error = -ENODEV;
1161                 goto err_unlock_ovs;
1162         }
1163         /* Check that the flow exists. */
1164         if (ufid_present)
1165                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1166         else
1167                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1168         if (unlikely(!flow)) {
1169                 error = -ENOENT;
1170                 goto err_unlock_ovs;
1171         }
1172
1173         /* Update actions, if present. */
1174         if (likely(acts)) {
1175                 old_acts = ovsl_dereference(flow->sf_acts);
1176                 rcu_assign_pointer(flow->sf_acts, acts);
1177
1178                 if (unlikely(reply)) {
1179                         error = ovs_flow_cmd_fill_info(flow,
1180                                                        ovs_header->dp_ifindex,
1181                                                        reply, info->snd_portid,
1182                                                        info->snd_seq, 0,
1183                                                        OVS_FLOW_CMD_NEW,
1184                                                        ufid_flags);
1185                         BUG_ON(error < 0);
1186                 }
1187         } else {
1188                 /* Could not alloc without acts before locking. */
1189                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1190                                                 info, OVS_FLOW_CMD_NEW, false,
1191                                                 ufid_flags);
1192
1193                 if (unlikely(IS_ERR(reply))) {
1194                         error = PTR_ERR(reply);
1195                         goto err_unlock_ovs;
1196                 }
1197         }
1198
1199         /* Clear stats. */
1200         if (a[OVS_FLOW_ATTR_CLEAR])
1201                 ovs_flow_stats_clear(flow);
1202         ovs_unlock();
1203
1204         if (reply)
1205                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1206         if (old_acts)
1207                 ovs_nla_free_flow_actions_rcu(old_acts);
1208
1209         return 0;
1210
1211 err_unlock_ovs:
1212         ovs_unlock();
1213         kfree_skb(reply);
1214 err_kfree_acts:
1215         ovs_nla_free_flow_actions(acts);
1216 error:
1217         return error;
1218 }
1219
1220 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1221 {
1222         struct nlattr **a = info->attrs;
1223         struct ovs_header *ovs_header = info->userhdr;
1224         struct net *net = sock_net(skb->sk);
1225         struct sw_flow_key key;
1226         struct sk_buff *reply;
1227         struct sw_flow *flow;
1228         struct datapath *dp;
1229         struct sw_flow_match match;
1230         struct sw_flow_id ufid;
1231         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1232         int err = 0;
1233         bool log = !a[OVS_FLOW_ATTR_PROBE];
1234         bool ufid_present;
1235
1236         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1237         if (a[OVS_FLOW_ATTR_KEY]) {
1238                 ovs_match_init(&match, &key, NULL);
1239                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1240                                         log);
1241         } else if (!ufid_present) {
1242                 OVS_NLERR(log,
1243                           "Flow get message rejected, Key attribute missing.");
1244                 err = -EINVAL;
1245         }
1246         if (err)
1247                 return err;
1248
1249         ovs_lock();
1250         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1251         if (!dp) {
1252                 err = -ENODEV;
1253                 goto unlock;
1254         }
1255
1256         if (ufid_present)
1257                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1258         else
1259                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1260         if (!flow) {
1261                 err = -ENOENT;
1262                 goto unlock;
1263         }
1264
1265         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1266                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1267         if (IS_ERR(reply)) {
1268                 err = PTR_ERR(reply);
1269                 goto unlock;
1270         }
1271
1272         ovs_unlock();
1273         return genlmsg_reply(reply, info);
1274 unlock:
1275         ovs_unlock();
1276         return err;
1277 }
1278
1279 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1280 {
1281         struct nlattr **a = info->attrs;
1282         struct ovs_header *ovs_header = info->userhdr;
1283         struct net *net = sock_net(skb->sk);
1284         struct sw_flow_key key;
1285         struct sk_buff *reply;
1286         struct sw_flow *flow = NULL;
1287         struct datapath *dp;
1288         struct sw_flow_match match;
1289         struct sw_flow_id ufid;
1290         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1291         int err;
1292         bool log = !a[OVS_FLOW_ATTR_PROBE];
1293         bool ufid_present;
1294
1295         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1296         if (a[OVS_FLOW_ATTR_KEY]) {
1297                 ovs_match_init(&match, &key, NULL);
1298                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1299                                         NULL, log);
1300                 if (unlikely(err))
1301                         return err;
1302         }
1303
1304         ovs_lock();
1305         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1306         if (unlikely(!dp)) {
1307                 err = -ENODEV;
1308                 goto unlock;
1309         }
1310
1311         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1312                 err = ovs_flow_tbl_flush(&dp->table);
1313                 goto unlock;
1314         }
1315
1316         if (ufid_present)
1317                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1318         else
1319                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1320         if (unlikely(!flow)) {
1321                 err = -ENOENT;
1322                 goto unlock;
1323         }
1324
1325         ovs_flow_tbl_remove(&dp->table, flow);
1326         ovs_unlock();
1327
1328         reply = ovs_flow_cmd_alloc_info(rcu_dereference_raw(flow->sf_acts),
1329                                         &flow->id, info, false, ufid_flags);
1330
1331         if (likely(reply)) {
1332                 if (likely(!IS_ERR(reply))) {
1333                         rcu_read_lock();        /*To keep RCU checker happy. */
1334                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1335                                                      reply, info->snd_portid,
1336                                                      info->snd_seq, 0,
1337                                                      OVS_FLOW_CMD_DEL,
1338                                                      ufid_flags);
1339                         rcu_read_unlock();
1340                         BUG_ON(err < 0);
1341                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1342                 } else {
1343                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1344                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1345
1346                 }
1347         }
1348
1349         ovs_flow_free(flow, true);
1350         return 0;
1351 unlock:
1352         ovs_unlock();
1353         return err;
1354 }
1355
1356 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1357 {
1358         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1359         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1360         struct table_instance *ti;
1361         struct datapath *dp;
1362         u32 ufid_flags;
1363         int err;
1364
1365         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1366                             OVS_FLOW_ATTR_MAX, flow_policy);
1367         if (err)
1368                 return err;
1369         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1370
1371         rcu_read_lock();
1372         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1373         if (!dp) {
1374                 rcu_read_unlock();
1375                 return -ENODEV;
1376         }
1377
1378         ti = rcu_dereference(dp->table.ti);
1379         for (;;) {
1380                 struct sw_flow *flow;
1381                 u32 bucket, obj;
1382
1383                 bucket = cb->args[0];
1384                 obj = cb->args[1];
1385                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1386                 if (!flow)
1387                         break;
1388
1389                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1390                                            NETLINK_CB(cb->skb).portid,
1391                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1392                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1393                         break;
1394
1395                 cb->args[0] = bucket;
1396                 cb->args[1] = obj;
1397         }
1398         rcu_read_unlock();
1399         return skb->len;
1400 }
1401
1402 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1403         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1404         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1405         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1406         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1407         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1408         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1409         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1410 };
1411
1412 static struct genl_ops dp_flow_genl_ops[] = {
1413         { .cmd = OVS_FLOW_CMD_NEW,
1414           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1415           .policy = flow_policy,
1416           .doit = ovs_flow_cmd_new
1417         },
1418         { .cmd = OVS_FLOW_CMD_DEL,
1419           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1420           .policy = flow_policy,
1421           .doit = ovs_flow_cmd_del
1422         },
1423         { .cmd = OVS_FLOW_CMD_GET,
1424           .flags = 0,               /* OK for unprivileged users. */
1425           .policy = flow_policy,
1426           .doit = ovs_flow_cmd_get,
1427           .dumpit = ovs_flow_cmd_dump
1428         },
1429         { .cmd = OVS_FLOW_CMD_SET,
1430           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431           .policy = flow_policy,
1432           .doit = ovs_flow_cmd_set,
1433         },
1434 };
1435
1436 static struct genl_family dp_flow_genl_family = {
1437         .id = GENL_ID_GENERATE,
1438         .hdrsize = sizeof(struct ovs_header),
1439         .name = OVS_FLOW_FAMILY,
1440         .version = OVS_FLOW_VERSION,
1441         .maxattr = OVS_FLOW_ATTR_MAX,
1442         .netnsok = true,
1443         .parallel_ops = true,
1444         .ops = dp_flow_genl_ops,
1445         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1446         .mcgrps = &ovs_dp_flow_multicast_group,
1447         .n_mcgrps = 1,
1448 };
1449
1450 static size_t ovs_dp_cmd_msg_size(void)
1451 {
1452         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1453
1454         msgsize += nla_total_size(IFNAMSIZ);
1455         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1456         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1457         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1458
1459         return msgsize;
1460 }
1461
1462 /* Called with ovs_mutex. */
1463 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1464                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1465 {
1466         struct ovs_header *ovs_header;
1467         struct ovs_dp_stats dp_stats;
1468         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1469         int err;
1470
1471         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1472                                    flags, cmd);
1473         if (!ovs_header)
1474                 goto error;
1475
1476         ovs_header->dp_ifindex = get_dpifindex(dp);
1477
1478         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1479         if (err)
1480                 goto nla_put_failure;
1481
1482         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1483         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1484                         &dp_stats))
1485                 goto nla_put_failure;
1486
1487         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1488                         sizeof(struct ovs_dp_megaflow_stats),
1489                         &dp_megaflow_stats))
1490                 goto nla_put_failure;
1491
1492         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1493                 goto nla_put_failure;
1494
1495         genlmsg_end(skb, ovs_header);
1496         return 0;
1497
1498 nla_put_failure:
1499         genlmsg_cancel(skb, ovs_header);
1500 error:
1501         return -EMSGSIZE;
1502 }
1503
1504 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1505 {
1506         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1507 }
1508
1509 /* Called with rcu_read_lock or ovs_mutex. */
1510 static struct datapath *lookup_datapath(struct net *net,
1511                                         const struct ovs_header *ovs_header,
1512                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1513 {
1514         struct datapath *dp;
1515
1516         if (!a[OVS_DP_ATTR_NAME])
1517                 dp = get_dp(net, ovs_header->dp_ifindex);
1518         else {
1519                 struct vport *vport;
1520
1521                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1522                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1523         }
1524         return dp ? dp : ERR_PTR(-ENODEV);
1525 }
1526
1527 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1528 {
1529         struct datapath *dp;
1530
1531         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1532         if (IS_ERR(dp))
1533                 return;
1534
1535         WARN(dp->user_features, "Dropping previously announced user features\n");
1536         dp->user_features = 0;
1537 }
1538
1539 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1540 {
1541         if (a[OVS_DP_ATTR_USER_FEATURES])
1542                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1543 }
1544
1545 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1546 {
1547         struct nlattr **a = info->attrs;
1548         struct vport_parms parms;
1549         struct sk_buff *reply;
1550         struct datapath *dp;
1551         struct vport *vport;
1552         struct ovs_net *ovs_net;
1553         int err, i;
1554
1555         err = -EINVAL;
1556         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1557                 goto err;
1558
1559         reply = ovs_dp_cmd_alloc_info(info);
1560         if (!reply)
1561                 return -ENOMEM;
1562
1563         err = -ENOMEM;
1564         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1565         if (dp == NULL)
1566                 goto err_free_reply;
1567
1568         ovs_dp_set_net(dp, sock_net(skb->sk));
1569
1570         /* Allocate table. */
1571         err = ovs_flow_tbl_init(&dp->table);
1572         if (err)
1573                 goto err_free_dp;
1574
1575         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1576         if (!dp->stats_percpu) {
1577                 err = -ENOMEM;
1578                 goto err_destroy_table;
1579         }
1580
1581         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1582                             GFP_KERNEL);
1583         if (!dp->ports) {
1584                 err = -ENOMEM;
1585                 goto err_destroy_percpu;
1586         }
1587
1588         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1589                 INIT_HLIST_HEAD(&dp->ports[i]);
1590
1591         /* Set up our datapath device. */
1592         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1593         parms.type = OVS_VPORT_TYPE_INTERNAL;
1594         parms.options = NULL;
1595         parms.dp = dp;
1596         parms.port_no = OVSP_LOCAL;
1597         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1598
1599         ovs_dp_change(dp, a);
1600
1601         /* So far only local changes have been made, now need the lock. */
1602         ovs_lock();
1603
1604         vport = new_vport(&parms);
1605         if (IS_ERR(vport)) {
1606                 err = PTR_ERR(vport);
1607                 if (err == -EBUSY)
1608                         err = -EEXIST;
1609
1610                 if (err == -EEXIST) {
1611                         /* An outdated user space instance that does not understand
1612                          * the concept of user_features has attempted to create a new
1613                          * datapath and is likely to reuse it. Drop all user features.
1614                          */
1615                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1616                                 ovs_dp_reset_user_features(skb, info);
1617                 }
1618
1619                 goto err_destroy_ports_array;
1620         }
1621
1622         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1623                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1624         BUG_ON(err < 0);
1625
1626         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1627         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1628
1629         ovs_unlock();
1630
1631         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1632         return 0;
1633
1634 err_destroy_ports_array:
1635         ovs_unlock();
1636         kfree(dp->ports);
1637 err_destroy_percpu:
1638         free_percpu(dp->stats_percpu);
1639 err_destroy_table:
1640         ovs_flow_tbl_destroy(&dp->table);
1641 err_free_dp:
1642         kfree(dp);
1643 err_free_reply:
1644         kfree_skb(reply);
1645 err:
1646         return err;
1647 }
1648
1649 /* Called with ovs_mutex. */
1650 static void __dp_destroy(struct datapath *dp)
1651 {
1652         int i;
1653
1654         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1655                 struct vport *vport;
1656                 struct hlist_node *n;
1657
1658                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1659                         if (vport->port_no != OVSP_LOCAL)
1660                                 ovs_dp_detach_port(vport);
1661         }
1662
1663         list_del_rcu(&dp->list_node);
1664
1665         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1666          * all ports in datapath are destroyed first before freeing datapath.
1667          */
1668         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1669
1670         /* RCU destroy the flow table */
1671         call_rcu(&dp->rcu, destroy_dp_rcu);
1672 }
1673
1674 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1675 {
1676         struct sk_buff *reply;
1677         struct datapath *dp;
1678         int err;
1679
1680         reply = ovs_dp_cmd_alloc_info(info);
1681         if (!reply)
1682                 return -ENOMEM;
1683
1684         ovs_lock();
1685         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1686         err = PTR_ERR(dp);
1687         if (IS_ERR(dp))
1688                 goto err_unlock_free;
1689
1690         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1691                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1692         BUG_ON(err < 0);
1693
1694         __dp_destroy(dp);
1695         ovs_unlock();
1696
1697         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1698         return 0;
1699
1700 err_unlock_free:
1701         ovs_unlock();
1702         kfree_skb(reply);
1703         return err;
1704 }
1705
1706 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1707 {
1708         struct sk_buff *reply;
1709         struct datapath *dp;
1710         int err;
1711
1712         reply = ovs_dp_cmd_alloc_info(info);
1713         if (!reply)
1714                 return -ENOMEM;
1715
1716         ovs_lock();
1717         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1718         err = PTR_ERR(dp);
1719         if (IS_ERR(dp))
1720                 goto err_unlock_free;
1721
1722         ovs_dp_change(dp, info->attrs);
1723
1724         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1725                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1726         BUG_ON(err < 0);
1727
1728         ovs_unlock();
1729
1730         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1731         return 0;
1732
1733 err_unlock_free:
1734         ovs_unlock();
1735         kfree_skb(reply);
1736         return err;
1737 }
1738
1739 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1740 {
1741         struct sk_buff *reply;
1742         struct datapath *dp;
1743         int err;
1744
1745         reply = ovs_dp_cmd_alloc_info(info);
1746         if (!reply)
1747                 return -ENOMEM;
1748
1749         ovs_lock();
1750         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1751         if (IS_ERR(dp)) {
1752                 err = PTR_ERR(dp);
1753                 goto err_unlock_free;
1754         }
1755         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1756                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1757         BUG_ON(err < 0);
1758         ovs_unlock();
1759
1760         return genlmsg_reply(reply, info);
1761
1762 err_unlock_free:
1763         ovs_unlock();
1764         kfree_skb(reply);
1765         return err;
1766 }
1767
1768 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1769 {
1770         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1771         struct datapath *dp;
1772         int skip = cb->args[0];
1773         int i = 0;
1774
1775         ovs_lock();
1776         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1777                 if (i >= skip &&
1778                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1779                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1780                                          OVS_DP_CMD_NEW) < 0)
1781                         break;
1782                 i++;
1783         }
1784         ovs_unlock();
1785
1786         cb->args[0] = i;
1787
1788         return skb->len;
1789 }
1790
1791 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1792         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1793         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1794         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1795 };
1796
1797 static struct genl_ops dp_datapath_genl_ops[] = {
1798         { .cmd = OVS_DP_CMD_NEW,
1799           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1800           .policy = datapath_policy,
1801           .doit = ovs_dp_cmd_new
1802         },
1803         { .cmd = OVS_DP_CMD_DEL,
1804           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1805           .policy = datapath_policy,
1806           .doit = ovs_dp_cmd_del
1807         },
1808         { .cmd = OVS_DP_CMD_GET,
1809           .flags = 0,               /* OK for unprivileged users. */
1810           .policy = datapath_policy,
1811           .doit = ovs_dp_cmd_get,
1812           .dumpit = ovs_dp_cmd_dump
1813         },
1814         { .cmd = OVS_DP_CMD_SET,
1815           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1816           .policy = datapath_policy,
1817           .doit = ovs_dp_cmd_set,
1818         },
1819 };
1820
1821 static struct genl_family dp_datapath_genl_family = {
1822         .id = GENL_ID_GENERATE,
1823         .hdrsize = sizeof(struct ovs_header),
1824         .name = OVS_DATAPATH_FAMILY,
1825         .version = OVS_DATAPATH_VERSION,
1826         .maxattr = OVS_DP_ATTR_MAX,
1827         .netnsok = true,
1828         .parallel_ops = true,
1829         .ops = dp_datapath_genl_ops,
1830         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1831         .mcgrps = &ovs_dp_datapath_multicast_group,
1832         .n_mcgrps = 1,
1833 };
1834
1835 /* Called with ovs_mutex or RCU read lock. */
1836 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1837                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1838 {
1839         struct ovs_header *ovs_header;
1840         struct ovs_vport_stats vport_stats;
1841         int err;
1842
1843         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1844                                  flags, cmd);
1845         if (!ovs_header)
1846                 return -EMSGSIZE;
1847
1848         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1849
1850         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1851             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1852             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1853                            ovs_vport_name(vport)))
1854                 goto nla_put_failure;
1855
1856         ovs_vport_get_stats(vport, &vport_stats);
1857         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1858                     &vport_stats))
1859                 goto nla_put_failure;
1860
1861         if (ovs_vport_get_upcall_portids(vport, skb))
1862                 goto nla_put_failure;
1863
1864         err = ovs_vport_get_options(vport, skb);
1865         if (err == -EMSGSIZE)
1866                 goto error;
1867
1868         genlmsg_end(skb, ovs_header);
1869         return 0;
1870
1871 nla_put_failure:
1872         err = -EMSGSIZE;
1873 error:
1874         genlmsg_cancel(skb, ovs_header);
1875         return err;
1876 }
1877
1878 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1879 {
1880         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1881 }
1882
1883 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1884 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1885                                          u32 seq, u8 cmd)
1886 {
1887         struct sk_buff *skb;
1888         int retval;
1889
1890         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1891         if (!skb)
1892                 return ERR_PTR(-ENOMEM);
1893
1894         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1895         BUG_ON(retval < 0);
1896
1897         return skb;
1898 }
1899
1900 /* Called with ovs_mutex or RCU read lock. */
1901 static struct vport *lookup_vport(struct net *net,
1902                                   const struct ovs_header *ovs_header,
1903                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1904 {
1905         struct datapath *dp;
1906         struct vport *vport;
1907
1908         if (a[OVS_VPORT_ATTR_NAME]) {
1909                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1910                 if (!vport)
1911                         return ERR_PTR(-ENODEV);
1912                 if (ovs_header->dp_ifindex &&
1913                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1914                         return ERR_PTR(-ENODEV);
1915                 return vport;
1916         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1917                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1918
1919                 if (port_no >= DP_MAX_PORTS)
1920                         return ERR_PTR(-EFBIG);
1921
1922                 dp = get_dp(net, ovs_header->dp_ifindex);
1923                 if (!dp)
1924                         return ERR_PTR(-ENODEV);
1925
1926                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1927                 if (!vport)
1928                         return ERR_PTR(-ENODEV);
1929                 return vport;
1930         } else
1931                 return ERR_PTR(-EINVAL);
1932 }
1933
1934 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1935 {
1936         struct nlattr **a = info->attrs;
1937         struct ovs_header *ovs_header = info->userhdr;
1938         struct vport_parms parms;
1939         struct sk_buff *reply;
1940         struct vport *vport;
1941         struct datapath *dp;
1942         u32 port_no;
1943         int err;
1944
1945         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1946             !a[OVS_VPORT_ATTR_UPCALL_PID])
1947                 return -EINVAL;
1948
1949         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1950                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1951         if (port_no >= DP_MAX_PORTS)
1952                 return -EFBIG;
1953
1954         reply = ovs_vport_cmd_alloc_info();
1955         if (!reply)
1956                 return -ENOMEM;
1957
1958         ovs_lock();
1959 restart:
1960         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1961         err = -ENODEV;
1962         if (!dp)
1963                 goto exit_unlock_free;
1964
1965         if (port_no) {
1966                 vport = ovs_vport_ovsl(dp, port_no);
1967                 err = -EBUSY;
1968                 if (vport)
1969                         goto exit_unlock_free;
1970         } else {
1971                 for (port_no = 1; ; port_no++) {
1972                         if (port_no >= DP_MAX_PORTS) {
1973                                 err = -EFBIG;
1974                                 goto exit_unlock_free;
1975                         }
1976                         vport = ovs_vport_ovsl(dp, port_no);
1977                         if (!vport)
1978                                 break;
1979                 }
1980         }
1981
1982         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1983         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1984         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1985         parms.dp = dp;
1986         parms.port_no = port_no;
1987         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1988
1989         vport = new_vport(&parms);
1990         err = PTR_ERR(vport);
1991         if (IS_ERR(vport)) {
1992                 if (err == -EAGAIN)
1993                         goto restart;
1994                 goto exit_unlock_free;
1995         }
1996
1997         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1998                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1999         BUG_ON(err < 0);
2000         ovs_unlock();
2001
2002         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2003         return 0;
2004
2005 exit_unlock_free:
2006         ovs_unlock();
2007         kfree_skb(reply);
2008         return err;
2009 }
2010
2011 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2012 {
2013         struct nlattr **a = info->attrs;
2014         struct sk_buff *reply;
2015         struct vport *vport;
2016         int err;
2017
2018         reply = ovs_vport_cmd_alloc_info();
2019         if (!reply)
2020                 return -ENOMEM;
2021
2022         ovs_lock();
2023         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2024         err = PTR_ERR(vport);
2025         if (IS_ERR(vport))
2026                 goto exit_unlock_free;
2027
2028         if (a[OVS_VPORT_ATTR_TYPE] &&
2029             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2030                 err = -EINVAL;
2031                 goto exit_unlock_free;
2032         }
2033
2034         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2035                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2036                 if (err)
2037                         goto exit_unlock_free;
2038         }
2039
2040         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2041                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2042
2043                 err = ovs_vport_set_upcall_portids(vport, ids);
2044                 if (err)
2045                         goto exit_unlock_free;
2046         }
2047
2048         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2049                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2050         BUG_ON(err < 0);
2051         ovs_unlock();
2052
2053         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2054         return 0;
2055
2056 exit_unlock_free:
2057         ovs_unlock();
2058         kfree_skb(reply);
2059         return err;
2060 }
2061
2062 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2063 {
2064         struct nlattr **a = info->attrs;
2065         struct sk_buff *reply;
2066         struct vport *vport;
2067         int err;
2068
2069         reply = ovs_vport_cmd_alloc_info();
2070         if (!reply)
2071                 return -ENOMEM;
2072
2073         ovs_lock();
2074         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2075         err = PTR_ERR(vport);
2076         if (IS_ERR(vport))
2077                 goto exit_unlock_free;
2078
2079         if (vport->port_no == OVSP_LOCAL) {
2080                 err = -EINVAL;
2081                 goto exit_unlock_free;
2082         }
2083
2084         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2085                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2086         BUG_ON(err < 0);
2087         ovs_dp_detach_port(vport);
2088         ovs_unlock();
2089
2090         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2091         return 0;
2092
2093 exit_unlock_free:
2094         ovs_unlock();
2095         kfree_skb(reply);
2096         return err;
2097 }
2098
2099 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2100 {
2101         struct nlattr **a = info->attrs;
2102         struct ovs_header *ovs_header = info->userhdr;
2103         struct sk_buff *reply;
2104         struct vport *vport;
2105         int err;
2106
2107         reply = ovs_vport_cmd_alloc_info();
2108         if (!reply)
2109                 return -ENOMEM;
2110
2111         rcu_read_lock();
2112         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2113         err = PTR_ERR(vport);
2114         if (IS_ERR(vport))
2115                 goto exit_unlock_free;
2116         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2117                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2118         BUG_ON(err < 0);
2119         rcu_read_unlock();
2120
2121         return genlmsg_reply(reply, info);
2122
2123 exit_unlock_free:
2124         rcu_read_unlock();
2125         kfree_skb(reply);
2126         return err;
2127 }
2128
2129 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2130 {
2131         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2132         struct datapath *dp;
2133         int bucket = cb->args[0], skip = cb->args[1];
2134         int i, j = 0;
2135
2136         rcu_read_lock();
2137         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2138         if (!dp) {
2139                 rcu_read_unlock();
2140                 return -ENODEV;
2141         }
2142         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2143                 struct vport *vport;
2144
2145                 j = 0;
2146                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2147                         if (j >= skip &&
2148                             ovs_vport_cmd_fill_info(vport, skb,
2149                                                     NETLINK_CB(cb->skb).portid,
2150                                                     cb->nlh->nlmsg_seq,
2151                                                     NLM_F_MULTI,
2152                                                     OVS_VPORT_CMD_NEW) < 0)
2153                                 goto out;
2154
2155                         j++;
2156                 }
2157                 skip = 0;
2158         }
2159 out:
2160         rcu_read_unlock();
2161
2162         cb->args[0] = i;
2163         cb->args[1] = j;
2164
2165         return skb->len;
2166 }
2167
2168 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2169         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2170         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2171         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2172         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2173         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2174         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2175 };
2176
2177 static struct genl_ops dp_vport_genl_ops[] = {
2178         { .cmd = OVS_VPORT_CMD_NEW,
2179           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2180           .policy = vport_policy,
2181           .doit = ovs_vport_cmd_new
2182         },
2183         { .cmd = OVS_VPORT_CMD_DEL,
2184           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2185           .policy = vport_policy,
2186           .doit = ovs_vport_cmd_del
2187         },
2188         { .cmd = OVS_VPORT_CMD_GET,
2189           .flags = 0,               /* OK for unprivileged users. */
2190           .policy = vport_policy,
2191           .doit = ovs_vport_cmd_get,
2192           .dumpit = ovs_vport_cmd_dump
2193         },
2194         { .cmd = OVS_VPORT_CMD_SET,
2195           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2196           .policy = vport_policy,
2197           .doit = ovs_vport_cmd_set,
2198         },
2199 };
2200
2201 struct genl_family dp_vport_genl_family = {
2202         .id = GENL_ID_GENERATE,
2203         .hdrsize = sizeof(struct ovs_header),
2204         .name = OVS_VPORT_FAMILY,
2205         .version = OVS_VPORT_VERSION,
2206         .maxattr = OVS_VPORT_ATTR_MAX,
2207         .netnsok = true,
2208         .parallel_ops = true,
2209         .ops = dp_vport_genl_ops,
2210         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2211         .mcgrps = &ovs_dp_vport_multicast_group,
2212         .n_mcgrps = 1,
2213 };
2214
2215 static struct genl_family *dp_genl_families[] = {
2216         &dp_datapath_genl_family,
2217         &dp_vport_genl_family,
2218         &dp_flow_genl_family,
2219         &dp_packet_genl_family,
2220 };
2221
2222 static void dp_unregister_genl(int n_families)
2223 {
2224         int i;
2225
2226         for (i = 0; i < n_families; i++)
2227                 genl_unregister_family(dp_genl_families[i]);
2228 }
2229
2230 static int dp_register_genl(void)
2231 {
2232         int err;
2233         int i;
2234
2235         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2236
2237                 err = genl_register_family(dp_genl_families[i]);
2238                 if (err)
2239                         goto error;
2240         }
2241
2242         return 0;
2243
2244 error:
2245         dp_unregister_genl(i);
2246         return err;
2247 }
2248
2249 static int __net_init ovs_init_net(struct net *net)
2250 {
2251         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2252
2253         INIT_LIST_HEAD(&ovs_net->dps);
2254         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2255         ovs_ct_init(net);
2256         return 0;
2257 }
2258
2259 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2260                                             struct list_head *head)
2261 {
2262         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2263         struct datapath *dp;
2264
2265         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2266                 int i;
2267
2268                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2269                         struct vport *vport;
2270
2271                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2272
2273                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2274                                         continue;
2275
2276                                 if (dev_net(vport->dev) == dnet)
2277                                         list_add(&vport->detach_list, head);
2278                         }
2279                 }
2280         }
2281 }
2282
2283 static void __net_exit ovs_exit_net(struct net *dnet)
2284 {
2285         struct datapath *dp, *dp_next;
2286         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2287         struct vport *vport, *vport_next;
2288         struct net *net;
2289         LIST_HEAD(head);
2290
2291         ovs_ct_exit(dnet);
2292         ovs_lock();
2293         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2294                 __dp_destroy(dp);
2295
2296         rtnl_lock();
2297         for_each_net(net)
2298                 list_vports_from_net(net, dnet, &head);
2299         rtnl_unlock();
2300
2301         /* Detach all vports from given namespace. */
2302         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2303                 list_del(&vport->detach_list);
2304                 ovs_dp_detach_port(vport);
2305         }
2306
2307         ovs_unlock();
2308
2309         cancel_work_sync(&ovs_net->dp_notify_work);
2310 }
2311
2312 static struct pernet_operations ovs_net_ops = {
2313         .init = ovs_init_net,
2314         .exit = ovs_exit_net,
2315         .id   = &ovs_net_id,
2316         .size = sizeof(struct ovs_net),
2317 };
2318
2319 DEFINE_COMPAT_PNET_REG_FUNC(device);
2320
2321 static int __init dp_init(void)
2322 {
2323         int err;
2324
2325         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2326
2327         pr_info("Open vSwitch switching datapath %s\n", VERSION);
2328
2329         err = compat_init();
2330         if (err)
2331                 goto error;
2332
2333         err = action_fifos_init();
2334         if (err)
2335                 goto error_compat_exit;
2336
2337         err = ovs_internal_dev_rtnl_link_register();
2338         if (err)
2339                 goto error_action_fifos_exit;
2340
2341         err = ovs_flow_init();
2342         if (err)
2343                 goto error_unreg_rtnl_link;
2344
2345         err = ovs_vport_init();
2346         if (err)
2347                 goto error_flow_exit;
2348
2349         err = register_pernet_device(&ovs_net_ops);
2350         if (err)
2351                 goto error_vport_exit;
2352
2353         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2354         if (err)
2355                 goto error_netns_exit;
2356
2357         err = ovs_netdev_init();
2358         if (err)
2359                 goto error_unreg_notifier;
2360
2361         err = dp_register_genl();
2362         if (err < 0)
2363                 goto error_unreg_netdev;
2364
2365         return 0;
2366
2367 error_unreg_netdev:
2368         ovs_netdev_exit();
2369 error_unreg_notifier:
2370         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2371 error_netns_exit:
2372         unregister_pernet_device(&ovs_net_ops);
2373 error_vport_exit:
2374         ovs_vport_exit();
2375 error_flow_exit:
2376         ovs_flow_exit();
2377 error_unreg_rtnl_link:
2378         ovs_internal_dev_rtnl_link_unregister();
2379 error_action_fifos_exit:
2380         action_fifos_exit();
2381 error_compat_exit:
2382         compat_exit();
2383 error:
2384         return err;
2385 }
2386
2387 static void dp_cleanup(void)
2388 {
2389         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2390         ovs_netdev_exit();
2391         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2392         unregister_pernet_device(&ovs_net_ops);
2393         rcu_barrier();
2394         ovs_vport_exit();
2395         ovs_flow_exit();
2396         ovs_internal_dev_rtnl_link_unregister();
2397         action_fifos_exit();
2398         compat_exit();
2399 }
2400
2401 module_init(dp_init);
2402 module_exit(dp_cleanup);
2403
2404 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2405 MODULE_LICENSE("GPL");
2406 MODULE_VERSION(VERSION);